?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Linux 4.13.16
[linux/fpc-iii.git] / net / sunrpc / svcsock.c
blobe18500151236ed3b162cebcf24966957e58be484
1 /*
2 * linux/net/sunrpc/svcsock.c
3 *
4 * These are the RPC server socket internals.
5 *
6 * The server scheduling algorithm does not always distribute the load
7 * evenly when servicing a single client. May need to modify the
8 * svc_xprt_enqueue procedure...
9 *
10 * TCP support is largely untested and may be a little slow. The problem
11 * is that we currently do two separate recvfrom's, one for the 4-byte
12 * record length, and the second for the actual record. This could possibly
13 * be improved by always reading a minimum size of around 100 bytes and
14 * tucking any superfluous bytes away in a temporary store. Still, that
15 * leaves write requests out in the rain. An alternative may be to peek at
16 * the first skb in the queue, and if it matches the next TCP sequence
17 * number, to extract the record marker. Yuck.
18 *
19 * Copyright (C) 1995, 1996 Olaf Kirch <okir@monad.swb.de>
20 */
21
22 #include <linux/kernel.h>
23 #include <linux/sched.h>
24 #include <linux/module.h>
25 #include <linux/errno.h>
26 #include <linux/fcntl.h>
27 #include <linux/net.h>
28 #include <linux/in.h>
29 #include <linux/inet.h>
30 #include <linux/udp.h>
31 #include <linux/tcp.h>
32 #include <linux/unistd.h>
33 #include <linux/slab.h>
34 #include <linux/netdevice.h>
35 #include <linux/skbuff.h>
36 #include <linux/file.h>
37 #include <linux/freezer.h>
38 #include <net/sock.h>
39 #include <net/checksum.h>
40 #include <net/ip.h>
41 #include <net/ipv6.h>
42 #include <net/udp.h>
43 #include <net/tcp.h>
44 #include <net/tcp_states.h>
45 #include <linux/uaccess.h>
46 #include <asm/ioctls.h>
47 #include <trace/events/skb.h>
48
49 #include <linux/sunrpc/types.h>
50 #include <linux/sunrpc/clnt.h>
51 #include <linux/sunrpc/xdr.h>
52 #include <linux/sunrpc/msg_prot.h>
53 #include <linux/sunrpc/svcsock.h>
54 #include <linux/sunrpc/stats.h>
55 #include <linux/sunrpc/xprt.h>
56
57 #include "sunrpc.h"
58
59 #define RPCDBG_FACILITY RPCDBG_SVCXPRT
60
61
62 static struct svc_sock *svc_setup_socket(struct svc_serv *, struct socket *,
63 int flags);
64 static int svc_udp_recvfrom(struct svc_rqst *);
65 static int svc_udp_sendto(struct svc_rqst *);
66 static void svc_sock_detach(struct svc_xprt *);
67 static void svc_tcp_sock_detach(struct svc_xprt *);
68 static void svc_sock_free(struct svc_xprt *);
69
70 static struct svc_xprt *svc_create_socket(struct svc_serv *, int,
71 struct net *, struct sockaddr *,
72 int, int);
73 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
74 static struct svc_xprt *svc_bc_create_socket(struct svc_serv *, int,
75 struct net *, struct sockaddr *,
76 int, int);
77 static void svc_bc_sock_free(struct svc_xprt *xprt);
78 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
79
80 #ifdef CONFIG_DEBUG_LOCK_ALLOC
81 static struct lock_class_key svc_key[2];
82 static struct lock_class_key svc_slock_key[2];
83
84 static void svc_reclassify_socket(struct socket *sock)
85 {
86 struct sock *sk = sock->sk;
87
88 if (WARN_ON_ONCE(!sock_allow_reclassification(sk)))
89 return;
90
91 switch (sk->sk_family) {
92 case AF_INET:
93 sock_lock_init_class_and_name(sk, "slock-AF_INET-NFSD",
94 &svc_slock_key[0],
95 "sk_xprt.xpt_lock-AF_INET-NFSD",
96 &svc_key[0]);
97 break;
98
99 case AF_INET6:
100 sock_lock_init_class_and_name(sk, "slock-AF_INET6-NFSD",
101 &svc_slock_key[1],
102 "sk_xprt.xpt_lock-AF_INET6-NFSD",
103 &svc_key[1]);
104 break;
105
106 default:
107 BUG();
108 }
109 }
110 #else
111 static void svc_reclassify_socket(struct socket *sock)
112 {
113 }
114 #endif
115
116 /*
117 * Release an skbuff after use
118 */
119 static void svc_release_skb(struct svc_rqst *rqstp)
120 {
121 struct sk_buff *skb = rqstp->rq_xprt_ctxt;
122
123 if (skb) {
124 struct svc_sock *svsk =
125 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
126 rqstp->rq_xprt_ctxt = NULL;
127
128 dprintk("svc: service %p, releasing skb %p\n", rqstp, skb);
129 skb_free_datagram_locked(svsk->sk_sk, skb);
130 }
131 }
132
133 static void svc_release_udp_skb(struct svc_rqst *rqstp)
134 {
135 struct sk_buff *skb = rqstp->rq_xprt_ctxt;
136
137 if (skb) {
138 rqstp->rq_xprt_ctxt = NULL;
139
140 dprintk("svc: service %p, releasing skb %p\n", rqstp, skb);
141 consume_skb(skb);
142 }
143 }
144
145 union svc_pktinfo_u {
146 struct in_pktinfo pkti;
147 struct in6_pktinfo pkti6;
148 };
149 #define SVC_PKTINFO_SPACE \
150 CMSG_SPACE(sizeof(union svc_pktinfo_u))
151
152 static void svc_set_cmsg_data(struct svc_rqst *rqstp, struct cmsghdr *cmh)
153 {
154 struct svc_sock *svsk =
155 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
156 switch (svsk->sk_sk->sk_family) {
157 case AF_INET: {
158 struct in_pktinfo *pki = CMSG_DATA(cmh);
159
160 cmh->cmsg_level = SOL_IP;
161 cmh->cmsg_type = IP_PKTINFO;
162 pki->ipi_ifindex = 0;
163 pki->ipi_spec_dst.s_addr =
164 svc_daddr_in(rqstp)->sin_addr.s_addr;
165 cmh->cmsg_len = CMSG_LEN(sizeof(*pki));
166 }
167 break;
168
169 case AF_INET6: {
170 struct in6_pktinfo *pki = CMSG_DATA(cmh);
171 struct sockaddr_in6 *daddr = svc_daddr_in6(rqstp);
172
173 cmh->cmsg_level = SOL_IPV6;
174 cmh->cmsg_type = IPV6_PKTINFO;
175 pki->ipi6_ifindex = daddr->sin6_scope_id;
176 pki->ipi6_addr = daddr->sin6_addr;
177 cmh->cmsg_len = CMSG_LEN(sizeof(*pki));
178 }
179 break;
180 }
181 }
182
183 /*
184 * send routine intended to be shared by the fore- and back-channel
185 */
186 int svc_send_common(struct socket *sock, struct xdr_buf *xdr,
187 struct page *headpage, unsigned long headoffset,
188 struct page *tailpage, unsigned long tailoffset)
189 {
190 int result;
191 int size;
192 struct page **ppage = xdr->pages;
193 size_t base = xdr->page_base;
194 unsigned int pglen = xdr->page_len;
195 unsigned int flags = MSG_MORE | MSG_SENDPAGE_NOTLAST;
196 int slen;
197 int len = 0;
198
199 slen = xdr->len;
200
201 /* send head */
202 if (slen == xdr->head[0].iov_len)
203 flags = 0;
204 len = kernel_sendpage(sock, headpage, headoffset,
205 xdr->head[0].iov_len, flags);
206 if (len != xdr->head[0].iov_len)
207 goto out;
208 slen -= xdr->head[0].iov_len;
209 if (slen == 0)
210 goto out;
211
212 /* send page data */
213 size = PAGE_SIZE - base < pglen ? PAGE_SIZE - base : pglen;
214 while (pglen > 0) {
215 if (slen == size)
216 flags = 0;
217 result = kernel_sendpage(sock, *ppage, base, size, flags);
218 if (result > 0)
219 len += result;
220 if (result != size)
221 goto out;
222 slen -= size;
223 pglen -= size;
224 size = PAGE_SIZE < pglen ? PAGE_SIZE : pglen;
225 base = 0;
226 ppage++;
227 }
228
229 /* send tail */
230 if (xdr->tail[0].iov_len) {
231 result = kernel_sendpage(sock, tailpage, tailoffset,
232 xdr->tail[0].iov_len, 0);
233 if (result > 0)
234 len += result;
235 }
236
237 out:
238 return len;
239 }
240
241
242 /*
243 * Generic sendto routine
244 */
245 static int svc_sendto(struct svc_rqst *rqstp, struct xdr_buf *xdr)
246 {
247 struct svc_sock *svsk =
248 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
249 struct socket *sock = svsk->sk_sock;
250 union {
251 struct cmsghdr hdr;
252 long all[SVC_PKTINFO_SPACE / sizeof(long)];
253 } buffer;
254 struct cmsghdr *cmh = &buffer.hdr;
255 int len = 0;
256 unsigned long tailoff;
257 unsigned long headoff;
258 RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
259
260 if (rqstp->rq_prot == IPPROTO_UDP) {
261 struct msghdr msg = {
262 .msg_name = &rqstp->rq_addr,
263 .msg_namelen = rqstp->rq_addrlen,
264 .msg_control = cmh,
265 .msg_controllen = sizeof(buffer),
266 .msg_flags = MSG_MORE,
267 };
268
269 svc_set_cmsg_data(rqstp, cmh);
270
271 if (sock_sendmsg(sock, &msg) < 0)
272 goto out;
273 }
274
275 tailoff = ((unsigned long)xdr->tail[0].iov_base) & (PAGE_SIZE-1);
276 headoff = 0;
277 len = svc_send_common(sock, xdr, rqstp->rq_respages[0], headoff,
278 rqstp->rq_respages[0], tailoff);
279
280 out:
281 dprintk("svc: socket %p sendto([%p %zu... ], %d) = %d (addr %s)\n",
282 svsk, xdr->head[0].iov_base, xdr->head[0].iov_len,
283 xdr->len, len, svc_print_addr(rqstp, buf, sizeof(buf)));
284
285 return len;
286 }
287
288 /*
289 * Report socket names for nfsdfs
290 */
291 static int svc_one_sock_name(struct svc_sock *svsk, char *buf, int remaining)
292 {
293 const struct sock *sk = svsk->sk_sk;
294 const char *proto_name = sk->sk_protocol == IPPROTO_UDP ?
295 "udp" : "tcp";
296 int len;
297
298 switch (sk->sk_family) {
299 case PF_INET:
300 len = snprintf(buf, remaining, "ipv4 %s %pI4 %d\n",
301 proto_name,
302 &inet_sk(sk)->inet_rcv_saddr,
303 inet_sk(sk)->inet_num);
304 break;
305 #if IS_ENABLED(CONFIG_IPV6)
306 case PF_INET6:
307 len = snprintf(buf, remaining, "ipv6 %s %pI6 %d\n",
308 proto_name,
309 &sk->sk_v6_rcv_saddr,
310 inet_sk(sk)->inet_num);
311 break;
312 #endif
313 default:
314 len = snprintf(buf, remaining, "*unknown-%d*\n",
315 sk->sk_family);
316 }
317
318 if (len >= remaining) {
319 *buf = '\0';
320 return -ENAMETOOLONG;
321 }
322 return len;
323 }
324
325 /*
326 * Generic recvfrom routine.
327 */
328 static int svc_recvfrom(struct svc_rqst *rqstp, struct kvec *iov, int nr,
329 int buflen)
330 {
331 struct svc_sock *svsk =
332 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
333 struct msghdr msg = {
334 .msg_flags = MSG_DONTWAIT,
335 };
336 int len;
337
338 rqstp->rq_xprt_hlen = 0;
339
340 clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
341 len = kernel_recvmsg(svsk->sk_sock, &msg, iov, nr, buflen,
342 msg.msg_flags);
343 /* If we read a full record, then assume there may be more
344 * data to read (stream based sockets only!)
345 */
346 if (len == buflen)
347 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
348
349 dprintk("svc: socket %p recvfrom(%p, %zu) = %d\n",
350 svsk, iov[0].iov_base, iov[0].iov_len, len);
351 return len;
352 }
353
354 static int svc_partial_recvfrom(struct svc_rqst *rqstp,
355 struct kvec *iov, int nr,
356 int buflen, unsigned int base)
357 {
358 size_t save_iovlen;
359 void *save_iovbase;
360 unsigned int i;
361 int ret;
362
363 if (base == 0)
364 return svc_recvfrom(rqstp, iov, nr, buflen);
365
366 for (i = 0; i < nr; i++) {
367 if (iov[i].iov_len > base)
368 break;
369 base -= iov[i].iov_len;
370 }
371 save_iovlen = iov[i].iov_len;
372 save_iovbase = iov[i].iov_base;
373 iov[i].iov_len -= base;
374 iov[i].iov_base += base;
375 ret = svc_recvfrom(rqstp, &iov[i], nr - i, buflen);
376 iov[i].iov_len = save_iovlen;
377 iov[i].iov_base = save_iovbase;
378 return ret;
379 }
380
381 /*
382 * Set socket snd and rcv buffer lengths
383 */
384 static void svc_sock_setbufsize(struct socket *sock, unsigned int snd,
385 unsigned int rcv)
386 {
387 #if 0
388 mm_segment_t oldfs;
389 oldfs = get_fs(); set_fs(KERNEL_DS);
390 sock_setsockopt(sock, SOL_SOCKET, SO_SNDBUF,
391 (char*)&snd, sizeof(snd));
392 sock_setsockopt(sock, SOL_SOCKET, SO_RCVBUF,
393 (char*)&rcv, sizeof(rcv));
394 #else
395 /* sock_setsockopt limits use to sysctl_?mem_max,
396 * which isn't acceptable. Until that is made conditional
397 * on not having CAP_SYS_RESOURCE or similar, we go direct...
398 * DaveM said I could!
399 */
400 lock_sock(sock->sk);
401 sock->sk->sk_sndbuf = snd * 2;
402 sock->sk->sk_rcvbuf = rcv * 2;
403 sock->sk->sk_write_space(sock->sk);
404 release_sock(sock->sk);
405 #endif
406 }
407
408 static int svc_sock_secure_port(struct svc_rqst *rqstp)
409 {
410 return svc_port_is_privileged(svc_addr(rqstp));
411 }
412
413 /*
414 * INET callback when data has been received on the socket.
415 */
416 static void svc_data_ready(struct sock *sk)
417 {
418 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
419
420 if (svsk) {
421 dprintk("svc: socket %p(inet %p), busy=%d\n",
422 svsk, sk,
423 test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags));
424
425 /* Refer to svc_setup_socket() for details. */
426 rmb();
427 svsk->sk_odata(sk);
428 if (!test_and_set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags))
429 svc_xprt_enqueue(&svsk->sk_xprt);
430 }
431 }
432
433 /*
434 * INET callback when space is newly available on the socket.
435 */
436 static void svc_write_space(struct sock *sk)
437 {
438 struct svc_sock *svsk = (struct svc_sock *)(sk->sk_user_data);
439
440 if (svsk) {
441 dprintk("svc: socket %p(inet %p), write_space busy=%d\n",
442 svsk, sk, test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags));
443
444 /* Refer to svc_setup_socket() for details. */
445 rmb();
446 svsk->sk_owspace(sk);
447 svc_xprt_enqueue(&svsk->sk_xprt);
448 }
449 }
450
451 static int svc_tcp_has_wspace(struct svc_xprt *xprt)
452 {
453 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
454
455 if (test_bit(XPT_LISTENER, &xprt->xpt_flags))
456 return 1;
457 return !test_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
458 }
459
460 static void svc_tcp_kill_temp_xprt(struct svc_xprt *xprt)
461 {
462 struct svc_sock *svsk;
463 struct socket *sock;
464 struct linger no_linger = {
465 .l_onoff = 1,
466 .l_linger = 0,
467 };
468
469 svsk = container_of(xprt, struct svc_sock, sk_xprt);
470 sock = svsk->sk_sock;
471 kernel_setsockopt(sock, SOL_SOCKET, SO_LINGER,
472 (char *)&no_linger, sizeof(no_linger));
473 }
474
475 /*
476 * See net/ipv6/ip_sockglue.c : ip_cmsg_recv_pktinfo
477 */
478 static int svc_udp_get_dest_address4(struct svc_rqst *rqstp,
479 struct cmsghdr *cmh)
480 {
481 struct in_pktinfo *pki = CMSG_DATA(cmh);
482 struct sockaddr_in *daddr = svc_daddr_in(rqstp);
483
484 if (cmh->cmsg_type != IP_PKTINFO)
485 return 0;
486
487 daddr->sin_family = AF_INET;
488 daddr->sin_addr.s_addr = pki->ipi_spec_dst.s_addr;
489 return 1;
490 }
491
492 /*
493 * See net/ipv6/datagram.c : ip6_datagram_recv_ctl
494 */
495 static int svc_udp_get_dest_address6(struct svc_rqst *rqstp,
496 struct cmsghdr *cmh)
497 {
498 struct in6_pktinfo *pki = CMSG_DATA(cmh);
499 struct sockaddr_in6 *daddr = svc_daddr_in6(rqstp);
500
501 if (cmh->cmsg_type != IPV6_PKTINFO)
502 return 0;
503
504 daddr->sin6_family = AF_INET6;
505 daddr->sin6_addr = pki->ipi6_addr;
506 daddr->sin6_scope_id = pki->ipi6_ifindex;
507 return 1;
508 }
509
510 /*
511 * Copy the UDP datagram's destination address to the rqstp structure.
512 * The 'destination' address in this case is the address to which the
513 * peer sent the datagram, i.e. our local address. For multihomed
514 * hosts, this can change from msg to msg. Note that only the IP
515 * address changes, the port number should remain the same.
516 */
517 static int svc_udp_get_dest_address(struct svc_rqst *rqstp,
518 struct cmsghdr *cmh)
519 {
520 switch (cmh->cmsg_level) {
521 case SOL_IP:
522 return svc_udp_get_dest_address4(rqstp, cmh);
523 case SOL_IPV6:
524 return svc_udp_get_dest_address6(rqstp, cmh);
525 }
526
527 return 0;
528 }
529
530 /*
531 * Receive a datagram from a UDP socket.
532 */
533 static int svc_udp_recvfrom(struct svc_rqst *rqstp)
534 {
535 struct svc_sock *svsk =
536 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
537 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
538 struct sk_buff *skb;
539 union {
540 struct cmsghdr hdr;
541 long all[SVC_PKTINFO_SPACE / sizeof(long)];
542 } buffer;
543 struct cmsghdr *cmh = &buffer.hdr;
544 struct msghdr msg = {
545 .msg_name = svc_addr(rqstp),
546 .msg_control = cmh,
547 .msg_controllen = sizeof(buffer),
548 .msg_flags = MSG_DONTWAIT,
549 };
550 size_t len;
551 int err;
552
553 if (test_and_clear_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags))
554 /* udp sockets need large rcvbuf as all pending
555 * requests are still in that buffer. sndbuf must
556 * also be large enough that there is enough space
557 * for one reply per thread. We count all threads
558 * rather than threads in a particular pool, which
559 * provides an upper bound on the number of threads
560 * which will access the socket.
561 */
562 svc_sock_setbufsize(svsk->sk_sock,
563 (serv->sv_nrthreads+3) * serv->sv_max_mesg,
564 (serv->sv_nrthreads+3) * serv->sv_max_mesg);
565
566 clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
567 skb = NULL;
568 err = kernel_recvmsg(svsk->sk_sock, &msg, NULL,
569 0, 0, MSG_PEEK | MSG_DONTWAIT);
570 if (err >= 0)
571 skb = skb_recv_udp(svsk->sk_sk, 0, 1, &err);
572
573 if (skb == NULL) {
574 if (err != -EAGAIN) {
575 /* possibly an icmp error */
576 dprintk("svc: recvfrom returned error %d\n", -err);
577 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
578 }
579 return 0;
580 }
581 len = svc_addr_len(svc_addr(rqstp));
582 rqstp->rq_addrlen = len;
583 if (skb->tstamp == 0) {
584 skb->tstamp = ktime_get_real();
585 /* Don't enable netstamp, sunrpc doesn't
586 need that much accuracy */
587 }
588 svsk->sk_sk->sk_stamp = skb->tstamp;
589 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); /* there may be more data... */
590
591 len = skb->len;
592 rqstp->rq_arg.len = len;
593
594 rqstp->rq_prot = IPPROTO_UDP;
595
596 if (!svc_udp_get_dest_address(rqstp, cmh)) {
597 net_warn_ratelimited("svc: received unknown control message %d/%d; dropping RPC reply datagram\n",
598 cmh->cmsg_level, cmh->cmsg_type);
599 goto out_free;
600 }
601 rqstp->rq_daddrlen = svc_addr_len(svc_daddr(rqstp));
602
603 if (skb_is_nonlinear(skb)) {
604 /* we have to copy */
605 local_bh_disable();
606 if (csum_partial_copy_to_xdr(&rqstp->rq_arg, skb)) {
607 local_bh_enable();
608 /* checksum error */
609 goto out_free;
610 }
611 local_bh_enable();
612 consume_skb(skb);
613 } else {
614 /* we can use it in-place */
615 rqstp->rq_arg.head[0].iov_base = skb->data;
616 rqstp->rq_arg.head[0].iov_len = len;
617 if (skb_checksum_complete(skb))
618 goto out_free;
619 rqstp->rq_xprt_ctxt = skb;
620 }
621
622 rqstp->rq_arg.page_base = 0;
623 if (len <= rqstp->rq_arg.head[0].iov_len) {
624 rqstp->rq_arg.head[0].iov_len = len;
625 rqstp->rq_arg.page_len = 0;
626 rqstp->rq_respages = rqstp->rq_pages+1;
627 } else {
628 rqstp->rq_arg.page_len = len - rqstp->rq_arg.head[0].iov_len;
629 rqstp->rq_respages = rqstp->rq_pages + 1 +
630 DIV_ROUND_UP(rqstp->rq_arg.page_len, PAGE_SIZE);
631 }
632 rqstp->rq_next_page = rqstp->rq_respages+1;
633
634 if (serv->sv_stats)
635 serv->sv_stats->netudpcnt++;
636
637 return len;
638 out_free:
639 kfree_skb(skb);
640 return 0;
641 }
642
643 static int
644 svc_udp_sendto(struct svc_rqst *rqstp)
645 {
646 int error;
647
648 error = svc_sendto(rqstp, &rqstp->rq_res);
649 if (error == -ECONNREFUSED)
650 /* ICMP error on earlier request. */
651 error = svc_sendto(rqstp, &rqstp->rq_res);
652
653 return error;
654 }
655
656 static void svc_udp_prep_reply_hdr(struct svc_rqst *rqstp)
657 {
658 }
659
660 static int svc_udp_has_wspace(struct svc_xprt *xprt)
661 {
662 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
663 struct svc_serv *serv = xprt->xpt_server;
664 unsigned long required;
665
666 /*
667 * Set the SOCK_NOSPACE flag before checking the available
668 * sock space.
669 */
670 set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
671 required = atomic_read(&svsk->sk_xprt.xpt_reserved) + serv->sv_max_mesg;
672 if (required*2 > sock_wspace(svsk->sk_sk))
673 return 0;
674 clear_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
675 return 1;
676 }
677
678 static struct svc_xprt *svc_udp_accept(struct svc_xprt *xprt)
679 {
680 BUG();
681 return NULL;
682 }
683
684 static void svc_udp_kill_temp_xprt(struct svc_xprt *xprt)
685 {
686 }
687
688 static struct svc_xprt *svc_udp_create(struct svc_serv *serv,
689 struct net *net,
690 struct sockaddr *sa, int salen,
691 int flags)
692 {
693 return svc_create_socket(serv, IPPROTO_UDP, net, sa, salen, flags);
694 }
695
696 static struct svc_xprt_ops svc_udp_ops = {
697 .xpo_create = svc_udp_create,
698 .xpo_recvfrom = svc_udp_recvfrom,
699 .xpo_sendto = svc_udp_sendto,
700 .xpo_release_rqst = svc_release_udp_skb,
701 .xpo_detach = svc_sock_detach,
702 .xpo_free = svc_sock_free,
703 .xpo_prep_reply_hdr = svc_udp_prep_reply_hdr,
704 .xpo_has_wspace = svc_udp_has_wspace,
705 .xpo_accept = svc_udp_accept,
706 .xpo_secure_port = svc_sock_secure_port,
707 .xpo_kill_temp_xprt = svc_udp_kill_temp_xprt,
708 };
709
710 static struct svc_xprt_class svc_udp_class = {
711 .xcl_name = "udp",
712 .xcl_owner = THIS_MODULE,
713 .xcl_ops = &svc_udp_ops,
714 .xcl_max_payload = RPCSVC_MAXPAYLOAD_UDP,
715 .xcl_ident = XPRT_TRANSPORT_UDP,
716 };
717
718 static void svc_udp_init(struct svc_sock *svsk, struct svc_serv *serv)
719 {
720 int err, level, optname, one = 1;
721
722 svc_xprt_init(sock_net(svsk->sk_sock->sk), &svc_udp_class,
723 &svsk->sk_xprt, serv);
724 clear_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags);
725 svsk->sk_sk->sk_data_ready = svc_data_ready;
726 svsk->sk_sk->sk_write_space = svc_write_space;
727
728 /* initialise setting must have enough space to
729 * receive and respond to one request.
730 * svc_udp_recvfrom will re-adjust if necessary
731 */
732 svc_sock_setbufsize(svsk->sk_sock,
733 3 * svsk->sk_xprt.xpt_server->sv_max_mesg,
734 3 * svsk->sk_xprt.xpt_server->sv_max_mesg);
735
736 /* data might have come in before data_ready set up */
737 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
738 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
739
740 /* make sure we get destination address info */
741 switch (svsk->sk_sk->sk_family) {
742 case AF_INET:
743 level = SOL_IP;
744 optname = IP_PKTINFO;
745 break;
746 case AF_INET6:
747 level = SOL_IPV6;
748 optname = IPV6_RECVPKTINFO;
749 break;
750 default:
751 BUG();
752 }
753 err = kernel_setsockopt(svsk->sk_sock, level, optname,
754 (char *)&one, sizeof(one));
755 dprintk("svc: kernel_setsockopt returned %d\n", err);
756 }
757
758 /*
759 * A data_ready event on a listening socket means there's a connection
760 * pending. Do not use state_change as a substitute for it.
761 */
762 static void svc_tcp_listen_data_ready(struct sock *sk)
763 {
764 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
765
766 dprintk("svc: socket %p TCP (listen) state change %d\n",
767 sk, sk->sk_state);
768
769 if (svsk) {
770 /* Refer to svc_setup_socket() for details. */
771 rmb();
772 svsk->sk_odata(sk);
773 }
774
775 /*
776 * This callback may called twice when a new connection
777 * is established as a child socket inherits everything
778 * from a parent LISTEN socket.
779 * 1) data_ready method of the parent socket will be called
780 * when one of child sockets become ESTABLISHED.
781 * 2) data_ready method of the child socket may be called
782 * when it receives data before the socket is accepted.
783 * In case of 2, we should ignore it silently.
784 */
785 if (sk->sk_state == TCP_LISTEN) {
786 if (svsk) {
787 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
788 svc_xprt_enqueue(&svsk->sk_xprt);
789 } else
790 printk("svc: socket %p: no user data\n", sk);
791 }
792 }
793
794 /*
795 * A state change on a connected socket means it's dying or dead.
796 */
797 static void svc_tcp_state_change(struct sock *sk)
798 {
799 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
800
801 dprintk("svc: socket %p TCP (connected) state change %d (svsk %p)\n",
802 sk, sk->sk_state, sk->sk_user_data);
803
804 if (!svsk)
805 printk("svc: socket %p: no user data\n", sk);
806 else {
807 /* Refer to svc_setup_socket() for details. */
808 rmb();
809 svsk->sk_ostate(sk);
810 if (sk->sk_state != TCP_ESTABLISHED) {
811 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
812 svc_xprt_enqueue(&svsk->sk_xprt);
813 }
814 }
815 }
816
817 /*
818 * Accept a TCP connection
819 */
820 static struct svc_xprt *svc_tcp_accept(struct svc_xprt *xprt)
821 {
822 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
823 struct sockaddr_storage addr;
824 struct sockaddr *sin = (struct sockaddr *) &addr;
825 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
826 struct socket *sock = svsk->sk_sock;
827 struct socket *newsock;
828 struct svc_sock *newsvsk;
829 int err, slen;
830 RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
831
832 dprintk("svc: tcp_accept %p sock %p\n", svsk, sock);
833 if (!sock)
834 return NULL;
835
836 clear_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
837 err = kernel_accept(sock, &newsock, O_NONBLOCK);
838 if (err < 0) {
839 if (err == -ENOMEM)
840 printk(KERN_WARNING "%s: no more sockets!\n",
841 serv->sv_name);
842 else if (err != -EAGAIN)
843 net_warn_ratelimited("%s: accept failed (err %d)!\n",
844 serv->sv_name, -err);
845 return NULL;
846 }
847 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
848
849 err = kernel_getpeername(newsock, sin, &slen);
850 if (err < 0) {
851 net_warn_ratelimited("%s: peername failed (err %d)!\n",
852 serv->sv_name, -err);
853 goto failed; /* aborted connection or whatever */
854 }
855
856 /* Ideally, we would want to reject connections from unauthorized
857 * hosts here, but when we get encryption, the IP of the host won't
858 * tell us anything. For now just warn about unpriv connections.
859 */
860 if (!svc_port_is_privileged(sin)) {
861 dprintk("%s: connect from unprivileged port: %s\n",
862 serv->sv_name,
863 __svc_print_addr(sin, buf, sizeof(buf)));
864 }
865 dprintk("%s: connect from %s\n", serv->sv_name,
866 __svc_print_addr(sin, buf, sizeof(buf)));
867
868 /* Reset the inherited callbacks before calling svc_setup_socket */
869 newsock->sk->sk_state_change = svsk->sk_ostate;
870 newsock->sk->sk_data_ready = svsk->sk_odata;
871 newsock->sk->sk_write_space = svsk->sk_owspace;
872
873 /* make sure that a write doesn't block forever when
874 * low on memory
875 */
876 newsock->sk->sk_sndtimeo = HZ*30;
877
878 newsvsk = svc_setup_socket(serv, newsock,
879 (SVC_SOCK_ANONYMOUS | SVC_SOCK_TEMPORARY));
880 if (IS_ERR(newsvsk))
881 goto failed;
882 svc_xprt_set_remote(&newsvsk->sk_xprt, sin, slen);
883 err = kernel_getsockname(newsock, sin, &slen);
884 if (unlikely(err < 0)) {
885 dprintk("svc_tcp_accept: kernel_getsockname error %d\n", -err);
886 slen = offsetof(struct sockaddr, sa_data);
887 }
888 svc_xprt_set_local(&newsvsk->sk_xprt, sin, slen);
889
890 if (sock_is_loopback(newsock->sk))
891 set_bit(XPT_LOCAL, &newsvsk->sk_xprt.xpt_flags);
892 else
893 clear_bit(XPT_LOCAL, &newsvsk->sk_xprt.xpt_flags);
894 if (serv->sv_stats)
895 serv->sv_stats->nettcpconn++;
896
897 return &newsvsk->sk_xprt;
898
899 failed:
900 sock_release(newsock);
901 return NULL;
902 }
903
904 static unsigned int svc_tcp_restore_pages(struct svc_sock *svsk, struct svc_rqst *rqstp)
905 {
906 unsigned int i, len, npages;
907
908 if (svsk->sk_datalen == 0)
909 return 0;
910 len = svsk->sk_datalen;
911 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
912 for (i = 0; i < npages; i++) {
913 if (rqstp->rq_pages[i] != NULL)
914 put_page(rqstp->rq_pages[i]);
915 BUG_ON(svsk->sk_pages[i] == NULL);
916 rqstp->rq_pages[i] = svsk->sk_pages[i];
917 svsk->sk_pages[i] = NULL;
918 }
919 rqstp->rq_arg.head[0].iov_base = page_address(rqstp->rq_pages[0]);
920 return len;
921 }
922
923 static void svc_tcp_save_pages(struct svc_sock *svsk, struct svc_rqst *rqstp)
924 {
925 unsigned int i, len, npages;
926
927 if (svsk->sk_datalen == 0)
928 return;
929 len = svsk->sk_datalen;
930 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
931 for (i = 0; i < npages; i++) {
932 svsk->sk_pages[i] = rqstp->rq_pages[i];
933 rqstp->rq_pages[i] = NULL;
934 }
935 }
936
937 static void svc_tcp_clear_pages(struct svc_sock *svsk)
938 {
939 unsigned int i, len, npages;
940
941 if (svsk->sk_datalen == 0)
942 goto out;
943 len = svsk->sk_datalen;
944 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
945 for (i = 0; i < npages; i++) {
946 if (svsk->sk_pages[i] == NULL) {
947 WARN_ON_ONCE(1);
948 continue;
949 }
950 put_page(svsk->sk_pages[i]);
951 svsk->sk_pages[i] = NULL;
952 }
953 out:
954 svsk->sk_tcplen = 0;
955 svsk->sk_datalen = 0;
956 }
957
958 /*
959 * Receive fragment record header.
960 * If we haven't gotten the record length yet, get the next four bytes.
961 */
962 static int svc_tcp_recv_record(struct svc_sock *svsk, struct svc_rqst *rqstp)
963 {
964 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
965 unsigned int want;
966 int len;
967
968 if (svsk->sk_tcplen < sizeof(rpc_fraghdr)) {
969 struct kvec iov;
970
971 want = sizeof(rpc_fraghdr) - svsk->sk_tcplen;
972 iov.iov_base = ((char *) &svsk->sk_reclen) + svsk->sk_tcplen;
973 iov.iov_len = want;
974 if ((len = svc_recvfrom(rqstp, &iov, 1, want)) < 0)
975 goto error;
976 svsk->sk_tcplen += len;
977
978 if (len < want) {
979 dprintk("svc: short recvfrom while reading record "
980 "length (%d of %d)\n", len, want);
981 return -EAGAIN;
982 }
983
984 dprintk("svc: TCP record, %d bytes\n", svc_sock_reclen(svsk));
985 if (svc_sock_reclen(svsk) + svsk->sk_datalen >
986 serv->sv_max_mesg) {
987 net_notice_ratelimited("RPC: fragment too large: %d\n",
988 svc_sock_reclen(svsk));
989 goto err_delete;
990 }
991 }
992
993 return svc_sock_reclen(svsk);
994 error:
995 dprintk("RPC: TCP recv_record got %d\n", len);
996 return len;
997 err_delete:
998 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
999 return -EAGAIN;
1000 }
1001
1002 static int receive_cb_reply(struct svc_sock *svsk, struct svc_rqst *rqstp)
1003 {
1004 struct rpc_xprt *bc_xprt = svsk->sk_xprt.xpt_bc_xprt;
1005 struct rpc_rqst *req = NULL;
1006 struct kvec *src, *dst;
1007 __be32 *p = (__be32 *)rqstp->rq_arg.head[0].iov_base;
1008 __be32 xid;
1009 __be32 calldir;
1010
1011 xid = *p++;
1012 calldir = *p;
1013
1014 if (!bc_xprt)
1015 return -EAGAIN;
1016 spin_lock_bh(&bc_xprt->transport_lock);
1017 req = xprt_lookup_rqst(bc_xprt, xid);
1018 if (!req)
1019 goto unlock_notfound;
1020
1021 memcpy(&req->rq_private_buf, &req->rq_rcv_buf, sizeof(struct xdr_buf));
1022 /*
1023 * XXX!: cheating for now! Only copying HEAD.
1024 * But we know this is good enough for now (in fact, for any
1025 * callback reply in the forseeable future).
1026 */
1027 dst = &req->rq_private_buf.head[0];
1028 src = &rqstp->rq_arg.head[0];
1029 if (dst->iov_len < src->iov_len)
1030 goto unlock_eagain; /* whatever; just giving up. */
1031 memcpy(dst->iov_base, src->iov_base, src->iov_len);
1032 xprt_complete_rqst(req->rq_task, rqstp->rq_arg.len);
1033 rqstp->rq_arg.len = 0;
1034 spin_unlock_bh(&bc_xprt->transport_lock);
1035 return 0;
1036 unlock_notfound:
1037 printk(KERN_NOTICE
1038 "%s: Got unrecognized reply: "
1039 "calldir 0x%x xpt_bc_xprt %p xid %08x\n",
1040 __func__, ntohl(calldir),
1041 bc_xprt, ntohl(xid));
1042 unlock_eagain:
1043 spin_unlock_bh(&bc_xprt->transport_lock);
1044 return -EAGAIN;
1045 }
1046
1047 static int copy_pages_to_kvecs(struct kvec *vec, struct page **pages, int len)
1048 {
1049 int i = 0;
1050 int t = 0;
1051
1052 while (t < len) {
1053 vec[i].iov_base = page_address(pages[i]);
1054 vec[i].iov_len = PAGE_SIZE;
1055 i++;
1056 t += PAGE_SIZE;
1057 }
1058 return i;
1059 }
1060
1061 static void svc_tcp_fragment_received(struct svc_sock *svsk)
1062 {
1063 /* If we have more data, signal svc_xprt_enqueue() to try again */
1064 dprintk("svc: TCP %s record (%d bytes)\n",
1065 svc_sock_final_rec(svsk) ? "final" : "nonfinal",
1066 svc_sock_reclen(svsk));
1067 svsk->sk_tcplen = 0;
1068 svsk->sk_reclen = 0;
1069 }
1070
1071 /*
1072 * Receive data from a TCP socket.
1073 */
1074 static int svc_tcp_recvfrom(struct svc_rqst *rqstp)
1075 {
1076 struct svc_sock *svsk =
1077 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
1078 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
1079 int len;
1080 struct kvec *vec;
1081 unsigned int want, base;
1082 __be32 *p;
1083 __be32 calldir;
1084 int pnum;
1085
1086 dprintk("svc: tcp_recv %p data %d conn %d close %d\n",
1087 svsk, test_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags),
1088 test_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags),
1089 test_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags));
1090
1091 len = svc_tcp_recv_record(svsk, rqstp);
1092 if (len < 0)
1093 goto error;
1094
1095 base = svc_tcp_restore_pages(svsk, rqstp);
1096 want = svc_sock_reclen(svsk) - (svsk->sk_tcplen - sizeof(rpc_fraghdr));
1097
1098 vec = rqstp->rq_vec;
1099
1100 pnum = copy_pages_to_kvecs(&vec[0], &rqstp->rq_pages[0],
1101 svsk->sk_datalen + want);
1102
1103 rqstp->rq_respages = &rqstp->rq_pages[pnum];
1104 rqstp->rq_next_page = rqstp->rq_respages + 1;
1105
1106 /* Now receive data */
1107 len = svc_partial_recvfrom(rqstp, vec, pnum, want, base);
1108 if (len >= 0) {
1109 svsk->sk_tcplen += len;
1110 svsk->sk_datalen += len;
1111 }
1112 if (len != want || !svc_sock_final_rec(svsk)) {
1113 svc_tcp_save_pages(svsk, rqstp);
1114 if (len < 0 && len != -EAGAIN)
1115 goto err_delete;
1116 if (len == want)
1117 svc_tcp_fragment_received(svsk);
1118 else
1119 dprintk("svc: incomplete TCP record (%d of %d)\n",
1120 (int)(svsk->sk_tcplen - sizeof(rpc_fraghdr)),
1121 svc_sock_reclen(svsk));
1122 goto err_noclose;
1123 }
1124
1125 if (svsk->sk_datalen < 8) {
1126 svsk->sk_datalen = 0;
1127 goto err_delete; /* client is nuts. */
1128 }
1129
1130 rqstp->rq_arg.len = svsk->sk_datalen;
1131 rqstp->rq_arg.page_base = 0;
1132 if (rqstp->rq_arg.len <= rqstp->rq_arg.head[0].iov_len) {
1133 rqstp->rq_arg.head[0].iov_len = rqstp->rq_arg.len;
1134 rqstp->rq_arg.page_len = 0;
1135 } else
1136 rqstp->rq_arg.page_len = rqstp->rq_arg.len - rqstp->rq_arg.head[0].iov_len;
1137
1138 rqstp->rq_xprt_ctxt = NULL;
1139 rqstp->rq_prot = IPPROTO_TCP;
1140 if (test_bit(XPT_LOCAL, &svsk->sk_xprt.xpt_flags))
1141 set_bit(RQ_LOCAL, &rqstp->rq_flags);
1142 else
1143 clear_bit(RQ_LOCAL, &rqstp->rq_flags);
1144
1145 p = (__be32 *)rqstp->rq_arg.head[0].iov_base;
1146 calldir = p[1];
1147 if (calldir)
1148 len = receive_cb_reply(svsk, rqstp);
1149
1150 /* Reset TCP read info */
1151 svsk->sk_datalen = 0;
1152 svc_tcp_fragment_received(svsk);
1153
1154 if (len < 0)
1155 goto error;
1156
1157 svc_xprt_copy_addrs(rqstp, &svsk->sk_xprt);
1158 if (serv->sv_stats)
1159 serv->sv_stats->nettcpcnt++;
1160
1161 return rqstp->rq_arg.len;
1162
1163 error:
1164 if (len != -EAGAIN)
1165 goto err_delete;
1166 dprintk("RPC: TCP recvfrom got EAGAIN\n");
1167 return 0;
1168 err_delete:
1169 printk(KERN_NOTICE "%s: recvfrom returned errno %d\n",
1170 svsk->sk_xprt.xpt_server->sv_name, -len);
1171 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
1172 err_noclose:
1173 return 0; /* record not complete */
1174 }
1175
1176 /*
1177 * Send out data on TCP socket.
1178 */
1179 static int svc_tcp_sendto(struct svc_rqst *rqstp)
1180 {
1181 struct xdr_buf *xbufp = &rqstp->rq_res;
1182 int sent;
1183 __be32 reclen;
1184
1185 /* Set up the first element of the reply kvec.
1186 * Any other kvecs that may be in use have been taken
1187 * care of by the server implementation itself.
1188 */
1189 reclen = htonl(0x80000000|((xbufp->len ) - 4));
1190 memcpy(xbufp->head[0].iov_base, &reclen, 4);
1191
1192 sent = svc_sendto(rqstp, &rqstp->rq_res);
1193 if (sent != xbufp->len) {
1194 printk(KERN_NOTICE
1195 "rpc-srv/tcp: %s: %s %d when sending %d bytes "
1196 "- shutting down socket\n",
1197 rqstp->rq_xprt->xpt_server->sv_name,
1198 (sent<0)?"got error":"sent only",
1199 sent, xbufp->len);
1200 set_bit(XPT_CLOSE, &rqstp->rq_xprt->xpt_flags);
1201 svc_xprt_enqueue(rqstp->rq_xprt);
1202 sent = -EAGAIN;
1203 }
1204 return sent;
1205 }
1206
1207 /*
1208 * Setup response header. TCP has a 4B record length field.
1209 */
1210 static void svc_tcp_prep_reply_hdr(struct svc_rqst *rqstp)
1211 {
1212 struct kvec *resv = &rqstp->rq_res.head[0];
1213
1214 /* tcp needs a space for the record length... */
1215 svc_putnl(resv, 0);
1216 }
1217
1218 static struct svc_xprt *svc_tcp_create(struct svc_serv *serv,
1219 struct net *net,
1220 struct sockaddr *sa, int salen,
1221 int flags)
1222 {
1223 return svc_create_socket(serv, IPPROTO_TCP, net, sa, salen, flags);
1224 }
1225
1226 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1227 static struct svc_xprt *svc_bc_create_socket(struct svc_serv *, int,
1228 struct net *, struct sockaddr *,
1229 int, int);
1230 static void svc_bc_sock_free(struct svc_xprt *xprt);
1231
1232 static struct svc_xprt *svc_bc_tcp_create(struct svc_serv *serv,
1233 struct net *net,
1234 struct sockaddr *sa, int salen,
1235 int flags)
1236 {
1237 return svc_bc_create_socket(serv, IPPROTO_TCP, net, sa, salen, flags);
1238 }
1239
1240 static void svc_bc_tcp_sock_detach(struct svc_xprt *xprt)
1241 {
1242 }
1243
1244 static struct svc_xprt_ops svc_tcp_bc_ops = {
1245 .xpo_create = svc_bc_tcp_create,
1246 .xpo_detach = svc_bc_tcp_sock_detach,
1247 .xpo_free = svc_bc_sock_free,
1248 .xpo_prep_reply_hdr = svc_tcp_prep_reply_hdr,
1249 .xpo_secure_port = svc_sock_secure_port,
1250 };
1251
1252 static struct svc_xprt_class svc_tcp_bc_class = {
1253 .xcl_name = "tcp-bc",
1254 .xcl_owner = THIS_MODULE,
1255 .xcl_ops = &svc_tcp_bc_ops,
1256 .xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP,
1257 };
1258
1259 static void svc_init_bc_xprt_sock(void)
1260 {
1261 svc_reg_xprt_class(&svc_tcp_bc_class);
1262 }
1263
1264 static void svc_cleanup_bc_xprt_sock(void)
1265 {
1266 svc_unreg_xprt_class(&svc_tcp_bc_class);
1267 }
1268 #else /* CONFIG_SUNRPC_BACKCHANNEL */
1269 static void svc_init_bc_xprt_sock(void)
1270 {
1271 }
1272
1273 static void svc_cleanup_bc_xprt_sock(void)
1274 {
1275 }
1276 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1277
1278 static struct svc_xprt_ops svc_tcp_ops = {
1279 .xpo_create = svc_tcp_create,
1280 .xpo_recvfrom = svc_tcp_recvfrom,
1281 .xpo_sendto = svc_tcp_sendto,
1282 .xpo_release_rqst = svc_release_skb,
1283 .xpo_detach = svc_tcp_sock_detach,
1284 .xpo_free = svc_sock_free,
1285 .xpo_prep_reply_hdr = svc_tcp_prep_reply_hdr,
1286 .xpo_has_wspace = svc_tcp_has_wspace,
1287 .xpo_accept = svc_tcp_accept,
1288 .xpo_secure_port = svc_sock_secure_port,
1289 .xpo_kill_temp_xprt = svc_tcp_kill_temp_xprt,
1290 };
1291
1292 static struct svc_xprt_class svc_tcp_class = {
1293 .xcl_name = "tcp",
1294 .xcl_owner = THIS_MODULE,
1295 .xcl_ops = &svc_tcp_ops,
1296 .xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP,
1297 .xcl_ident = XPRT_TRANSPORT_TCP,
1298 };
1299
1300 void svc_init_xprt_sock(void)
1301 {
1302 svc_reg_xprt_class(&svc_tcp_class);
1303 svc_reg_xprt_class(&svc_udp_class);
1304 svc_init_bc_xprt_sock();
1305 }
1306
1307 void svc_cleanup_xprt_sock(void)
1308 {
1309 svc_unreg_xprt_class(&svc_tcp_class);
1310 svc_unreg_xprt_class(&svc_udp_class);
1311 svc_cleanup_bc_xprt_sock();
1312 }
1313
1314 static void svc_tcp_init(struct svc_sock *svsk, struct svc_serv *serv)
1315 {
1316 struct sock *sk = svsk->sk_sk;
1317
1318 svc_xprt_init(sock_net(svsk->sk_sock->sk), &svc_tcp_class,
1319 &svsk->sk_xprt, serv);
1320 set_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags);
1321 set_bit(XPT_CONG_CTRL, &svsk->sk_xprt.xpt_flags);
1322 if (sk->sk_state == TCP_LISTEN) {
1323 dprintk("setting up TCP socket for listening\n");
1324 set_bit(XPT_LISTENER, &svsk->sk_xprt.xpt_flags);
1325 sk->sk_data_ready = svc_tcp_listen_data_ready;
1326 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
1327 } else {
1328 dprintk("setting up TCP socket for reading\n");
1329 sk->sk_state_change = svc_tcp_state_change;
1330 sk->sk_data_ready = svc_data_ready;
1331 sk->sk_write_space = svc_write_space;
1332
1333 svsk->sk_reclen = 0;
1334 svsk->sk_tcplen = 0;
1335 svsk->sk_datalen = 0;
1336 memset(&svsk->sk_pages[0], 0, sizeof(svsk->sk_pages));
1337
1338 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
1339
1340 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
1341 switch (sk->sk_state) {
1342 case TCP_SYN_RECV:
1343 case TCP_ESTABLISHED:
1344 break;
1345 default:
1346 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
1347 }
1348 }
1349 }
1350
1351 void svc_sock_update_bufs(struct svc_serv *serv)
1352 {
1353 /*
1354 * The number of server threads has changed. Update
1355 * rcvbuf and sndbuf accordingly on all sockets
1356 */
1357 struct svc_sock *svsk;
1358
1359 spin_lock_bh(&serv->sv_lock);
1360 list_for_each_entry(svsk, &serv->sv_permsocks, sk_xprt.xpt_list)
1361 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
1362 spin_unlock_bh(&serv->sv_lock);
1363 }
1364 EXPORT_SYMBOL_GPL(svc_sock_update_bufs);
1365
1366 /*
1367 * Initialize socket for RPC use and create svc_sock struct
1368 */
1369 static struct svc_sock *svc_setup_socket(struct svc_serv *serv,
1370 struct socket *sock,
1371 int flags)
1372 {
1373 struct svc_sock *svsk;
1374 struct sock *inet;
1375 int pmap_register = !(flags & SVC_SOCK_ANONYMOUS);
1376 int err = 0;
1377
1378 dprintk("svc: svc_setup_socket %p\n", sock);
1379 svsk = kzalloc(sizeof(*svsk), GFP_KERNEL);
1380 if (!svsk)
1381 return ERR_PTR(-ENOMEM);
1382
1383 inet = sock->sk;
1384
1385 /* Register socket with portmapper */
1386 if (pmap_register)
1387 err = svc_register(serv, sock_net(sock->sk), inet->sk_family,
1388 inet->sk_protocol,
1389 ntohs(inet_sk(inet)->inet_sport));
1390
1391 if (err < 0) {
1392 kfree(svsk);
1393 return ERR_PTR(err);
1394 }
1395
1396 svsk->sk_sock = sock;
1397 svsk->sk_sk = inet;
1398 svsk->sk_ostate = inet->sk_state_change;
1399 svsk->sk_odata = inet->sk_data_ready;
1400 svsk->sk_owspace = inet->sk_write_space;
1401 /*
1402 * This barrier is necessary in order to prevent race condition
1403 * with svc_data_ready(), svc_listen_data_ready() and others
1404 * when calling callbacks above.
1405 */
1406 wmb();
1407 inet->sk_user_data = svsk;
1408
1409 /* Initialize the socket */
1410 if (sock->type == SOCK_DGRAM)
1411 svc_udp_init(svsk, serv);
1412 else
1413 svc_tcp_init(svsk, serv);
1414
1415 dprintk("svc: svc_setup_socket created %p (inet %p), "
1416 "listen %d close %d\n",
1417 svsk, svsk->sk_sk,
1418 test_bit(XPT_LISTENER, &svsk->sk_xprt.xpt_flags),
1419 test_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags));
1420
1421 return svsk;
1422 }
1423
1424 bool svc_alien_sock(struct net *net, int fd)
1425 {
1426 int err;
1427 struct socket *sock = sockfd_lookup(fd, &err);
1428 bool ret = false;
1429
1430 if (!sock)
1431 goto out;
1432 if (sock_net(sock->sk) != net)
1433 ret = true;
1434 sockfd_put(sock);
1435 out:
1436 return ret;
1437 }
1438 EXPORT_SYMBOL_GPL(svc_alien_sock);
1439
1440 /**
1441 * svc_addsock - add a listener socket to an RPC service
1442 * @serv: pointer to RPC service to which to add a new listener
1443 * @fd: file descriptor of the new listener
1444 * @name_return: pointer to buffer to fill in with name of listener
1445 * @len: size of the buffer
1446 *
1447 * Fills in socket name and returns positive length of name if successful.
1448 * Name is terminated with '\n'. On error, returns a negative errno
1449 * value.
1450 */
1451 int svc_addsock(struct svc_serv *serv, const int fd, char *name_return,
1452 const size_t len)
1453 {
1454 int err = 0;
1455 struct socket *so = sockfd_lookup(fd, &err);
1456 struct svc_sock *svsk = NULL;
1457 struct sockaddr_storage addr;
1458 struct sockaddr *sin = (struct sockaddr *)&addr;
1459 int salen;
1460
1461 if (!so)
1462 return err;
1463 err = -EAFNOSUPPORT;
1464 if ((so->sk->sk_family != PF_INET) && (so->sk->sk_family != PF_INET6))
1465 goto out;
1466 err = -EPROTONOSUPPORT;
1467 if (so->sk->sk_protocol != IPPROTO_TCP &&
1468 so->sk->sk_protocol != IPPROTO_UDP)
1469 goto out;
1470 err = -EISCONN;
1471 if (so->state > SS_UNCONNECTED)
1472 goto out;
1473 err = -ENOENT;
1474 if (!try_module_get(THIS_MODULE))
1475 goto out;
1476 svsk = svc_setup_socket(serv, so, SVC_SOCK_DEFAULTS);
1477 if (IS_ERR(svsk)) {
1478 module_put(THIS_MODULE);
1479 err = PTR_ERR(svsk);
1480 goto out;
1481 }
1482 if (kernel_getsockname(svsk->sk_sock, sin, &salen) == 0)
1483 svc_xprt_set_local(&svsk->sk_xprt, sin, salen);
1484 svc_add_new_perm_xprt(serv, &svsk->sk_xprt);
1485 return svc_one_sock_name(svsk, name_return, len);
1486 out:
1487 sockfd_put(so);
1488 return err;
1489 }
1490 EXPORT_SYMBOL_GPL(svc_addsock);
1491
1492 /*
1493 * Create socket for RPC service.
1494 */
1495 static struct svc_xprt *svc_create_socket(struct svc_serv *serv,
1496 int protocol,
1497 struct net *net,
1498 struct sockaddr *sin, int len,
1499 int flags)
1500 {
1501 struct svc_sock *svsk;
1502 struct socket *sock;
1503 int error;
1504 int type;
1505 struct sockaddr_storage addr;
1506 struct sockaddr *newsin = (struct sockaddr *)&addr;
1507 int newlen;
1508 int family;
1509 int val;
1510 RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
1511
1512 dprintk("svc: svc_create_socket(%s, %d, %s)\n",
1513 serv->sv_program->pg_name, protocol,
1514 __svc_print_addr(sin, buf, sizeof(buf)));
1515
1516 if (protocol != IPPROTO_UDP && protocol != IPPROTO_TCP) {
1517 printk(KERN_WARNING "svc: only UDP and TCP "
1518 "sockets supported\n");
1519 return ERR_PTR(-EINVAL);
1520 }
1521
1522 type = (protocol == IPPROTO_UDP)? SOCK_DGRAM : SOCK_STREAM;
1523 switch (sin->sa_family) {
1524 case AF_INET6:
1525 family = PF_INET6;
1526 break;
1527 case AF_INET:
1528 family = PF_INET;
1529 break;
1530 default:
1531 return ERR_PTR(-EINVAL);
1532 }
1533
1534 error = __sock_create(net, family, type, protocol, &sock, 1);
1535 if (error < 0)
1536 return ERR_PTR(error);
1537
1538 svc_reclassify_socket(sock);
1539
1540 /*
1541 * If this is an PF_INET6 listener, we want to avoid
1542 * getting requests from IPv4 remotes. Those should
1543 * be shunted to a PF_INET listener via rpcbind.
1544 */
1545 val = 1;
1546 if (family == PF_INET6)
1547 kernel_setsockopt(sock, SOL_IPV6, IPV6_V6ONLY,
1548 (char *)&val, sizeof(val));
1549
1550 if (type == SOCK_STREAM)
1551 sock->sk->sk_reuse = SK_CAN_REUSE; /* allow address reuse */
1552 error = kernel_bind(sock, sin, len);
1553 if (error < 0)
1554 goto bummer;
1555
1556 newlen = len;
1557 error = kernel_getsockname(sock, newsin, &newlen);
1558 if (error < 0)
1559 goto bummer;
1560
1561 if (protocol == IPPROTO_TCP) {
1562 if ((error = kernel_listen(sock, 64)) < 0)
1563 goto bummer;
1564 }
1565
1566 svsk = svc_setup_socket(serv, sock, flags);
1567 if (IS_ERR(svsk)) {
1568 error = PTR_ERR(svsk);
1569 goto bummer;
1570 }
1571 svc_xprt_set_local(&svsk->sk_xprt, newsin, newlen);
1572 return (struct svc_xprt *)svsk;
1573 bummer:
1574 dprintk("svc: svc_create_socket error = %d\n", -error);
1575 sock_release(sock);
1576 return ERR_PTR(error);
1577 }
1578
1579 /*
1580 * Detach the svc_sock from the socket so that no
1581 * more callbacks occur.
1582 */
1583 static void svc_sock_detach(struct svc_xprt *xprt)
1584 {
1585 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1586 struct sock *sk = svsk->sk_sk;
1587
1588 dprintk("svc: svc_sock_detach(%p)\n", svsk);
1589
1590 /* put back the old socket callbacks */
1591 lock_sock(sk);
1592 sk->sk_state_change = svsk->sk_ostate;
1593 sk->sk_data_ready = svsk->sk_odata;
1594 sk->sk_write_space = svsk->sk_owspace;
1595 sk->sk_user_data = NULL;
1596 release_sock(sk);
1597 }
1598
1599 /*
1600 * Disconnect the socket, and reset the callbacks
1601 */
1602 static void svc_tcp_sock_detach(struct svc_xprt *xprt)
1603 {
1604 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1605
1606 dprintk("svc: svc_tcp_sock_detach(%p)\n", svsk);
1607
1608 svc_sock_detach(xprt);
1609
1610 if (!test_bit(XPT_LISTENER, &xprt->xpt_flags)) {
1611 svc_tcp_clear_pages(svsk);
1612 kernel_sock_shutdown(svsk->sk_sock, SHUT_RDWR);
1613 }
1614 }
1615
1616 /*
1617 * Free the svc_sock's socket resources and the svc_sock itself.
1618 */
1619 static void svc_sock_free(struct svc_xprt *xprt)
1620 {
1621 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1622 dprintk("svc: svc_sock_free(%p)\n", svsk);
1623
1624 if (svsk->sk_sock->file)
1625 sockfd_put(svsk->sk_sock);
1626 else
1627 sock_release(svsk->sk_sock);
1628 kfree(svsk);
1629 }
1630
1631 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1632 /*
1633 * Create a back channel svc_xprt which shares the fore channel socket.
1634 */
1635 static struct svc_xprt *svc_bc_create_socket(struct svc_serv *serv,
1636 int protocol,
1637 struct net *net,
1638 struct sockaddr *sin, int len,
1639 int flags)
1640 {
1641 struct svc_sock *svsk;
1642 struct svc_xprt *xprt;
1643
1644 if (protocol != IPPROTO_TCP) {
1645 printk(KERN_WARNING "svc: only TCP sockets"
1646 " supported on shared back channel\n");
1647 return ERR_PTR(-EINVAL);
1648 }
1649
1650 svsk = kzalloc(sizeof(*svsk), GFP_KERNEL);
1651 if (!svsk)
1652 return ERR_PTR(-ENOMEM);
1653
1654 xprt = &svsk->sk_xprt;
1655 svc_xprt_init(net, &svc_tcp_bc_class, xprt, serv);
1656 set_bit(XPT_CONG_CTRL, &svsk->sk_xprt.xpt_flags);
1657
1658 serv->sv_bc_xprt = xprt;
1659
1660 return xprt;
1661 }
1662
1663 /*
1664 * Free a back channel svc_sock.
1665 */
1666 static void svc_bc_sock_free(struct svc_xprt *xprt)
1667 {
1668 if (xprt)
1669 kfree(container_of(xprt, struct svc_sock, sk_xprt));
1670 }
1671 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
Linux with added FPC-III support